Backdoor Defense on GTSRB BadNets (test)
100Attack Success RateNo Defense
Evaluation results
| Method | Links | ||
|---|---|---|---|
| No DefenseBackbone=ResNet-182021.12 | 100 | 95.17 | |
| No DefenseBackbone=VGG2021.12 | 99.6 | 97.03 | |
| ANPBackbone=VGG, Available Data Size (n)=50002022.07 | 99.06 | 5.17 | |
| GSBackbone=ResNet-182021.12 | 57.73 | 89.46 | |
| Neural CleanseBackbone=ResNet-182021.12 | 43.1 | 90.44 | |
| ANPBackbone=VGG, Available Data Size (n)=5002022.07 | 6.35 | 97.35 | |
| Fine-PruningBackbone=VGG2021.12 | 6.16 | 96.26 | |
| Neural CleanseBackbone=VGG2021.12 | 4.76 | 96.69 | |
| AWMBackbone=VGG, Available Data Size (n)=50002022.07 | 4.35 | 97.32 | |
| AWMBackbone=VGG, Available Data Size (n)=5002022.07 | 4.31 | 98.9 | |
| AWMBackbone=Resnet-18, Available Data Size (n)=One-shot2022.07 | 3.67 | 96.91 | |
| ANPBackbone=Resnet-18, Available Data Size (n)=50002022.07 | 3.56 | — | |
| AWMBackbone=Resnet-18, Available Data Size (n)=50002022.07 | 3.53 | 99.33 | |
| IBAUBackbone=Resnet-18, Available Data Size (n)=50002022.07 | 3.47 | 99.23 | |
| GSBackbone=VGG2021.12 | 1.11 | 96.63 | |
| Fine-PruningBackbone=ResNet-182021.12 | 0.79 | 17.03 | |
| ShapPruningBackbone=VGG, Trigger Type=Synthesized2021.12 | 0.49 | 96.94 | |
| ShapPruningBackbone=ResNet-18, Trigger Type=Synthesized2021.12 | 0.48 | 92.25 | |
| ShapPruningBackbone=VGG, Trigger Type=Original2021.12 | 0.46 | 97.16 | |
| IBAUBackbone=VGG, Available Data Size (n)=50002022.07 | 0.37 | — | |
| IBAUBackbone=VGG, Available Data Size (n)=5002022.07 | 0.32 | 97.02 | |
| ShapPruningBackbone=ResNet-18, Trigger Type=Original2021.12 | 0.2 | 92.71 |